nolo-cli 0.1.19 → 0.1.20

package/ai/agent/streamAgentChatTurn.ts

@@ -180,6 +180,17 @@ const extractAgentRunUserText = (userInput: string | any[]) => {
         .trim();
 };
 
+const requiresServerResolvedProviderConfig = (agentConfig: Agent): boolean => {
+    if (agentConfig.useServerProxy !== true) return false;
+
+    const provider = String(agentConfig.provider ?? "").toLowerCase();
+    const apiSource = String((agentConfig as any).apiSource ?? "").toLowerCase();
+    const isCustomProvider = provider === "custom" || apiSource === "custom";
+    if (!isCustomProvider) return false;
+
+    return !agentConfig.customProviderUrl?.trim() || !agentConfig.apiKey?.trim();
+};
+
 const hasAgentRunUserInputContent = (userInput: string | any[]) => {
     if (typeof userInput === "string") {
         return userInput.trim().length > 0;
@@ -612,12 +623,18 @@ export const streamAgentChatTurnHandler = async (
             typeof args.serverBase === "string" && args.serverBase.trim()
                 ? args.serverBase.trim()
                 : null;
+        const currentServer = selectCurrentServer(state);
         const declaredRuntimeServerBase = extractAgentRuntimeServerBase(agentConfig);
-        const requestedServerBase = explicitServerBase ?? declaredRuntimeServerBase;
+        const needsServerResolvedProviderConfig =
+            requiresServerResolvedProviderConfig(agentConfig);
+        const requestedServerBase =
+            explicitServerBase ??
+            declaredRuntimeServerBase ??
+            (needsServerResolvedProviderConfig ? currentServer : null);
         const normalizedRequestedServerBase =
             requestedServerBase && normalizeServerOrigin(requestedServerBase);
         const normalizedCurrentServer = normalizeServerOrigin(
-            selectCurrentServer(state),
+            currentServer,
         );
         const canAutoRouteRemotely =
             !Array.isArray(userInput) &&
@@ -626,6 +643,7 @@ export const streamAgentChatTurnHandler = async (
             !runtimeOptions?.imageConfigOverride;
         if (requestedServerBase && canAutoRouteRemotely) {
             if (
+                !needsServerResolvedProviderConfig &&
                 normalizedRequestedServerBase &&
                 normalizedCurrentServer &&
                 normalizedRequestedServerBase === normalizedCurrentServer

package/ai/agent/streamAgentChatTurnUtils.ts

@@ -28,7 +28,12 @@ import {
     selectCurrentUserBalance,
     selectUserId,
 } from "auth/authSlice";
-import { getModelPricing, getPrices, getFinalPrice } from "ai/llm/getPricing";
+import {
+    getModelPricing,
+    getPrices,
+    getFinalPrice,
+    hasExplicitAgentPricing,
+} from "ai/llm/getPricing";
 import {
     buildStaticUserPolicyContext,
     resolveSpaceContextPreloadPlan,
@@ -52,6 +57,35 @@ import {
 import { buildRecentAppToolMemory } from "./appWorkingMemory";
 import type { AgentRuntimeOptions } from "./types";
 
+const BROWSER_UNAVAILABLE_CORE_TOOLS = new Set([
+    "queryModelUsage",
+    "createDialogGoal",
+    "getDialogGoal",
+    "completeDialogGoal",
+    "createScheduledTask",
+    "notifyUser",
+]);
+
+const getRuntimeCoreTools = (): string[] => {
+    if (typeof window === "undefined") {
+        return TOOL_PACKS.CORE;
+    }
+    return TOOL_PACKS.CORE.filter(
+        (toolName) => !BROWSER_UNAVAILABLE_CORE_TOOLS.has(toolName),
+    );
+};
+
+const isInlineVisualArtifactAgent = (agentConfig: Agent): boolean => {
+    const tags = Array.isArray((agentConfig as any).tags)
+        ? ((agentConfig as any).tags as unknown[])
+        : [];
+    return tags.some(
+        (tag) =>
+            typeof tag === "string" &&
+            ["inline-artifact", "streaming-ui"].includes(tag)
+    );
+};
+
 /**
  * 估算单条 OpenAI 消息的 token 数（包括 tool_calls）。
  */
@@ -353,11 +387,11 @@ export const validateAccessAndBalance = (
 
     const serverPrices = getModelPricing(agentConfig.provider || "", agentConfig.model);
 
-    if (!serverPrices) {
+    if (!serverPrices && !hasExplicitAgentPricing(agentConfig)) {
         return "无法获取模型定价信息，请稍后重试。";
     }
 
-    const prices = getPrices(agentConfig, serverPrices);
+    const prices = getPrices(agentConfig, serverPrices ?? null);
     const maxPrice = getFinalPrice(prices);
 
     if (userBalance < maxPrice) {
@@ -677,11 +711,17 @@ export const mergeAgentToolsWithRuntime = (
     );
     const enhancedTools = new Set<string>([
         ...baseTools,
-        ...TOOL_PACKS.CORE,
-        ...(baseTools.length > 0 ? TOOL_PACKS.LIGHT_WEB : []),
-        ...requiredSkillTools,
-        ...canonicalizeToolNames(referencedTools),
-        ...canonicalizeToolNames(mentionedTools),
+        ...(isInlineVisualArtifactAgent(agentConfig) ? [] : getRuntimeCoreTools()),
+        ...(baseTools.length > 0 && !isInlineVisualArtifactAgent(agentConfig)
+            ? TOOL_PACKS.LIGHT_WEB
+            : []),
+        ...(isInlineVisualArtifactAgent(agentConfig) ? [] : requiredSkillTools),
+        ...(isInlineVisualArtifactAgent(agentConfig)
+            ? []
+            : canonicalizeToolNames(referencedTools)),
+        ...(isInlineVisualArtifactAgent(agentConfig)
+            ? []
+            : canonicalizeToolNames(mentionedTools)),
     ]);
 
     // Intelligence: If user explicitly added ANY browser tool, auto-inject the FULL browser pack
@@ -690,18 +730,22 @@ export const mergeAgentToolsWithRuntime = (
         TOOL_PACKS.FULL_BROWSER.forEach((t) => enhancedTools.add(t));
     }
 
-    const extraTools = canonicalizeToolNames(runtimeOptions?.extraTools ?? []);
+    const extraTools = isInlineVisualArtifactAgent(agentConfig)
+        ? []
+        : canonicalizeToolNames(runtimeOptions?.extraTools ?? []);
     for (const t of extraTools) {
         enhancedTools.add(t);
     }
 
-    const viewMode = state ? selectViewMode(state) : "categories";
-    if (viewMode === "all") {
-        enhancedTools.delete("search_workspace");
-        enhancedTools.add("search_all_spaces");
-    } else {
-        enhancedTools.delete("search_all_spaces");
-        enhancedTools.add("search_workspace");
+    if (!isInlineVisualArtifactAgent(agentConfig)) {
+        const viewMode = state ? selectViewMode(state) : "categories";
+        if (viewMode === "all") {
+            enhancedTools.delete("search_workspace");
+            enhancedTools.add("search_all_spaces");
+        } else {
+            enhancedTools.delete("search_all_spaces");
+            enhancedTools.add("search_workspace");
+        }
     }
 
     return {

package/ai/chat/accumulateToolCallChunks.ts

@@ -6,13 +6,14 @@
  * - 不再过滤特殊标记，保持原样透传
  */
 
+
 export interface ToolCallChunk {
   index?: number;
   id?: string;
   type?: "function";
   function?: {
     name?: string;
-    arguments?: string;
+    arguments?: string | object;
   };
 }
 
@@ -22,7 +23,7 @@ export interface AccumulatedToolCall {
   type: "function";
   function: {
     name: string;
-    arguments: string;
+    arguments: string | object;
   };
 }
 
@@ -39,9 +40,12 @@ export function accumulateToolCallChunks(
     if (index !== undefined) {
       // 确保数组长度足够覆盖 index
       while (out.length <= index) {
-        // 先占位，后续必须填充 id/type/function 才能成为有效的 AccumulatedToolCall
-        // 这里暂时断言为空对象，等待后续逻辑填充完整
-        out.push({} as AccumulatedToolCall);
+        // 先占位，后续填充。初始化所有必需字段以防空指针。
+        out.push({
+          id: "",
+          type: "function",
+          function: { name: "", arguments: "" },
+        });
       }
 
       const cur = out[index];
@@ -53,7 +57,20 @@ export function accumulateToolCallChunks(
 
       if (fn) {
         if (fn.name) cur.function.name += fn.name;
-        if (fn.arguments) cur.function.arguments += fn.arguments;
+
+        if (fn.arguments) {
+          if (typeof fn.arguments === "string") {
+            // 字符串增量：追加
+            const currentArgs =
+              typeof cur.function.arguments === "string"
+                ? cur.function.arguments
+                : "";
+            cur.function.arguments = currentArgs + fn.arguments;
+          } else {
+            // 对象全量：覆盖（非标流直接给 final object）
+            cur.function.arguments = fn.arguments;
+          }
+        }
       }
       continue;
     }
@@ -68,14 +85,21 @@ export function accumulateToolCallChunks(
         targetIndex = out.findIndex((c) => c.id === id);
       } else if (out.length > 0) {
         // 如果没有 ID，默认追加到最后一个（假设顺序性）
-        // 注意：这是兜底逻辑，OpenAI 规范通常会带 index 或 id
         targetIndex = out.length - 1;
       }
 
       if (targetIndex >= 0) {
         const target = out[targetIndex];
         if (fn.name) target.function.name += fn.name; // 追加
-        if (fn.arguments) target.function.arguments += fn.arguments; // 追加
+
+        if (fn.arguments) {
+          if (typeof fn.arguments === "string") {
+            const currentArgs = typeof target.function.arguments === "string" ? target.function.arguments : "";
+            target.function.arguments = currentArgs + fn.arguments;
+          } else {
+            target.function.arguments = fn.arguments;
+          }
+        }
       } else if (id) {
         // 是新的调用
         const newCall: AccumulatedToolCall = {
@@ -83,9 +107,16 @@ export function accumulateToolCallChunks(
           type: type || "function",
           function: {
             name: fn.name || "",
-            arguments: fn.arguments || ""
+            arguments: fn.arguments || (!fn.arguments && typeof fn.arguments === 'object' ? {} : "") // Initial empty value based on type? Or just default string
           },
         };
+        // For arguments, if it's object, use it. If undefined, use "".
+        if (fn.arguments) {
+          newCall.function.arguments = fn.arguments;
+        } else {
+          newCall.function.arguments = "";
+        }
+
         out.push(newCall);
       }
     }

package/ai/chat/parseApiError.ts

@@ -28,6 +28,9 @@ export async function parseApiError(response: Response): Promise<string> {
       if (isContextOverflow(errorMessage) || isContextOverflow(errorBody) || errorCode === "UPSTREAM_400") {
         return "上下文过长：本轮消息或工具结果太大。请缩小范围，或先读取更小片段后再继续。";
       }
+      if (errorCode === "MISSING_PROVIDER_API_KEY") {
+        return truncateErrorMessage(errorMessage);
+      }
       if (errorMessage && errorMessage !== defaultMessage) {
         return `请求参数错误: ${truncateErrorMessage(errorMessage)}`;
       }

package/ai/chat/sendOpenAICompletionsRequest.native.ts

@@ -12,7 +12,7 @@ import {
     messageStreaming,
 } from "chat/messages/messageSlice";
 import { handleToolCalls } from "chat/messages/toolThunks";
-import { MessageContentPart, OpenAITextContent } from "chat/messages/types";
+import { CompletionFinishReason, CompletionUsage, MessageContentPart, OpenAITextContent } from "chat/messages/types";
 import { selectCurrentServer } from "app/settings/settingSlice";
 import { getApiEndpoint } from "ai/llm/providers";
 import { createDialogMessageKeyAndId } from "database/keys";
@@ -23,7 +23,7 @@ import { performSSEFetchRequest } from "./fetchUtils";
 import { createSSEParser } from "./parseMultilineSSE";
 import { parseApiError } from "./parseApiError";
 import { updateTotalUsage } from "./updateTotalUsage";
-import { accumulateToolCallChunks, AccumulatedToolCall, ToolCallChunk } from "./accumulateToolCallChunks";
+import { accumulateToolCallChunks, AccumulatedToolCall } from "./accumulateToolCallChunks";
 import { prepareTools } from "../tools/prepareTools";
 
 import { getModelInfo } from "ai/llm/getModelContextWindow";
@@ -65,14 +65,14 @@ type AssistantToolCall = {
 /** 单次流式请求过程中的全部中间状态（显式 state） */
 type StreamState = {
     contentBuffer: MessageContentPart[];
-    totalUsage: any | null;
+    totalUsage: CompletionUsage | null;
     accumulatedToolCalls: AccumulatedToolCall[];
     reasoningBuffer: string;
     assistantToolCalls?: AssistantToolCall[];
     hasHandedOff: boolean;
     hasProcessedToolCalls: boolean;
     alreadyFinalized: boolean;
-    finishReason: string | null;
+    finishReason: CompletionFinishReason | null;
 };
 
 type FinalizeContext = {
@@ -100,12 +100,13 @@ type StreamCompletionContext = {
     agentConfig: any;
 };
 
+
 /** 单轮调用后返回给 Agent Loop 的元信息 */
 export type CompletionMeta = {
     hasToolCalls: boolean;
     hasPendingInteraction: boolean;
     hasHandedOff: boolean;
-    finishReason: string | null;
+    finishReason: CompletionFinishReason | null;
     usage?: any;
 };
 
@@ -222,11 +223,11 @@ function applyDelta(
         next = {
             ...next,
             accumulatedToolCalls: accumulated,
-            assistantToolCalls: accumulated.map((call) => ({
+            assistantToolCalls: accumulated.map((call: AccumulatedToolCall) => ({
                 id: call.id,
                 type: "function",
                 function: {
-                    name: call.function?.name,
+                    name: call.function?.name || '', // Ensure name is string
                     arguments:
                         typeof call.function?.arguments === "string"
                             ? call.function.arguments
@@ -417,6 +418,18 @@ export const sendOpenAICompletionsRequest = async ({
         return "Unknown error";
     };
 
+    const formatStreamErrorMessage = (data: any): string => {
+        const rawMessage = getStreamErrorMessage(data);
+        if (
+            /prohibited|violation|terms\s+of\s+service|content\s+policy|safety/i.test(
+                rawMessage
+            )
+        ) {
+            return "当前模型服务商拒绝了这次请求。你可以稍后重试，或切换到其他模型继续。";
+        }
+        return rawMessage;
+    };
+
     const { dispatch, getState, signal: thunkSignal } = thunkApi;
 
     const dialogId = extractCustomId(dialogKey);
@@ -460,7 +473,7 @@ export const sendOpenAICompletionsRequest = async ({
 
     let hasHandedOffOverall = false;
     let hasPendingInteractionOverall = false;
-    let lastFinishReason: string | null = null;
+    let lastFinishReason: CompletionFinishReason = null;
 
     const buildMeta = (): CompletionMeta => ({
         hasToolCalls:
@@ -551,7 +564,7 @@ export const sendOpenAICompletionsRequest = async ({
                             }
 
                             if (data.error) {
-                                const errorMsg = `Error: ${getStreamErrorMessage(data)}`;
+                                const errorMsg = `Error: ${formatStreamErrorMessage(data)}`;
                                 streamState = {
                                     ...streamState,
                                     contentBuffer: appendTextChunk(
@@ -585,7 +598,7 @@ export const sendOpenAICompletionsRequest = async ({
                                 dialogId,
                             });
 
-                            const finishReason = choice.finish_reason;
+                            const finishReason = choice.finish_reason as CompletionFinishReason;
                             if (finishReason) {
                                 lastFinishReason = finishReason;
                                 streamState.finishReason = finishReason;

package/ai/chat/sendOpenAICompletionsRequest.ts

@@ -97,6 +97,18 @@ function getStreamErrorMessage(data: any): string {
   return "Unknown error";
 }
 
+function formatStreamErrorMessage(data: any): string {
+  const rawMessage = getStreamErrorMessage(data);
+  if (
+    /prohibited|violation|terms\s+of\s+service|content\s+policy|safety/i.test(
+      rawMessage
+    )
+  ) {
+    return "当前模型服务商拒绝了这次请求。你可以稍后重试，或切换到其他模型继续。";
+  }
+  return rawMessage;
+}
+
 /** 单次流式请求过程中的全部中间状态（显式 state） */
 type StreamState = {
   contentBuffer: MessageContentPart[];
@@ -607,7 +619,7 @@ export const sendOpenAICompletionsRequest = async ({
           }
 
           if (data.error) {
-            const errorMsg = `Error: ${getStreamErrorMessage(data)}`;
+            const errorMsg = `Error: ${formatStreamErrorMessage(data)}`;
             streamState = {
               ...streamState,
               contentBuffer: appendTextChunk(

package/ai/chat/updateTotalUsage.ts

@@ -1,5 +1,7 @@
 // 文件路径: ai/chat/updateTotalUsage.ts
 
+import { CompletionUsage } from "chat/messages/types";
+
 /**
  * ✨ 新增辅助函数 ✨
  * 根据新的数据块更新累积的 token 使用量。
@@ -7,26 +9,41 @@
  * @param newUsageChunk - 从流中收到的新 usage 数据块。
  * @returns 更新后的 usage 对象。
  */
-export function updateTotalUsage(currentUsage: any, newUsageChunk: any): any {
+export function updateTotalUsage(
+  currentUsage: CompletionUsage | null,
+  newUsageChunk: Partial<CompletionUsage>
+): CompletionUsage | null {
   if (!newUsageChunk) {
     return currentUsage;
   }
 
   // 如果是第一次接收，直接克隆新数据块
   if (!currentUsage) {
-    return { ...newUsageChunk };
+    return {
+      completion_tokens: 0,
+      prompt_tokens: 0,
+      total_tokens: 0,
+      ...newUsageChunk,
+    } as CompletionUsage;
   }
 
   // 否则，在现有基础上进行累加或更新
-  const updatedUsage = { ...currentUsage };
+  const updatedUsage: CompletionUsage = { ...currentUsage };
 
   // === token 相关 ===
-  updatedUsage.completion_tokens =
-    newUsageChunk.completion_tokens ?? updatedUsage.completion_tokens;
-  updatedUsage.prompt_tokens =
-    newUsageChunk.prompt_tokens ?? updatedUsage.prompt_tokens;
-  updatedUsage.total_tokens =
-    newUsageChunk.total_tokens ?? updatedUsage.total_tokens;
+  // 这里假设 usage 是累积的还是单次的？
+  // 如果是 OpenAI 流式 output，usage 通常只在最后一次 chunk 发送完整的统计值 (除了 Azure 等可能变体)。
+  // 如果是这种情况，我们直接覆盖即可。
+  // 但如果服务端分片发送增量（比较少见但有可能），则需要累加。
+  // 原代码逻辑是：
+  // updatedUsage.completion_tokens = newUsageChunk.completion_tokens ?? updatedUsage.completion_tokens;
+  // 这意味着如果有新值就覆盖，没新值保持原样。这适合 "最后一次发送完整值" 的场景。
+
+  // 保持原逻辑：覆盖
+  if (newUsageChunk.completion_tokens !== undefined) updatedUsage.completion_tokens = newUsageChunk.completion_tokens;
+  if (newUsageChunk.prompt_tokens !== undefined) updatedUsage.prompt_tokens = newUsageChunk.prompt_tokens;
+  if (newUsageChunk.total_tokens !== undefined) updatedUsage.total_tokens = newUsageChunk.total_tokens;
+
 
   if (newUsageChunk.prompt_tokens_details) {
     updatedUsage.prompt_tokens_details = {

package/ai/llm/deepinfra.ts

@@ -1,5 +1,20 @@
 // ai/llm/deepinfra.ts
+import { DEEPINFRA_KIMI_FALLBACK_MODEL } from "./kimi";
+
 export const deepinfraModels = [
+  {
+    name: DEEPINFRA_KIMI_FALLBACK_MODEL,
+    displayName: "MoonshotAI: Kimi K2.6 (DeepInfra)",
+    hasVision: true,
+    price: {
+      input: 0.6 * 8,
+      output: 2.5 * 8,
+      inputCacheHit: 0.06 * 8,
+    },
+    maxOutputTokens: 262144,
+    contextWindow: 262144,
+    supportsTool: true,
+  },
   {
     name: "moonshotai/Kimi-K2.5",
     displayName: "MoonshotAI: Kimi K2.5 (DeepInfra)",
@@ -25,4 +40,40 @@ export const deepinfraModels = [
     contextWindow: 202752,
     supportsTool: true,
   },
+  {
+    name: "anthropic/claude-haiku-4-5",
+    displayName: "Anthropic: Claude Haiku 4.5 (DeepInfra)",
+    hasVision: true,
+    price: {
+      input: 1 * 9,
+      output: 5 * 9,
+    },
+    contextWindow: 195000,
+    maxOutputTokens: 4092,
+    supportsTool: false,
+  },
+  {
+    name: "anthropic/claude-sonnet-4-6",
+    displayName: "Anthropic: Claude Sonnet 4.6 (DeepInfra)",
+    hasVision: true,
+    price: {
+      input: 3 * 9,
+      output: 15 * 9,
+    },
+    contextWindow: 976000,
+    maxOutputTokens: 4092,
+    supportsTool: false,
+  },
+  {
+    name: "anthropic/claude-opus-4-7",
+    displayName: "Anthropic: Claude Opus 4.7 (DeepInfra)",
+    hasVision: true,
+    price: {
+      input: 5 * 9,
+      output: 25 * 9,
+    },
+    contextWindow: 976000,
+    maxOutputTokens: 4092,
+    supportsTool: false,
+  },
 ];

package/ai/llm/getPricing.ts

@@ -70,6 +70,12 @@ export const getModelPricingForModel = (
   };
 };
 
+export const hasExplicitAgentPricing = (config: any): boolean =>
+  [config?.inputPrice, config?.outputPrice].some(
+    (value) =>
+      typeof value === "number" && Number.isFinite(value) && value > 0
+  );
+
 export const getPrices = (config: any, serverPrices: any): Prices => ({
   cybotInput: Number(config?.inputPrice ?? 0),
   cybotOutput: Number(config?.outputPrice ?? 0),

package/ai/llm/kimi.ts

@@ -1,6 +1,8 @@
 export const FIREWORKS_KIMI_LATEST_MODEL = "accounts/fireworks/models/kimi-latest";
 export const FIREWORKS_KIMI_CURRENT_MODEL = "accounts/fireworks/models/kimi-k2p6";
+export const DEEPINFRA_KIMI_FALLBACK_MODEL = "moonshotai/Kimi-K2.6";
 export const OPENROUTER_KIMI_FALLBACK_MODEL = "moonshotai/kimi-k2.6";
+export const KIMI_PLATFORM_FALLBACK_STATUSES = [402, 429, 500, 502, 503, 504];
 
 export const isFireworksKimiModel = (model?: string | null): boolean =>
   model === FIREWORKS_KIMI_LATEST_MODEL || model === FIREWORKS_KIMI_CURRENT_MODEL;